semi-automated topology adjustment of partial …

SEMI-AUTOMATED TOPOLOGY ADJUSTMENT OF PARTIAL WALL GEOMETRY

M. Bassier ∗

Dept. of Civil Engineering, TC Construction - GeomaticsKU Leuven - Faculty of Engineering Technology

Ghent, Belgium(maarten.bassier,maarten.vergauwen)@kuleuven.be

KEY WORDS: Building Information Modeling, Topology, Walls, Building, Point Clouds

ABSTRACT:

The reconstruction of Building Information Modeling objects for as-built modeling is currently the subject of ongoing research.A popular method is to extract building information from point cloud data to create a set of parametric objects. The automationof this process is highly desired by the industry but is currently hindered by occlusions, clutter and the complexity of the buildinggeometry. To create an as-built BIM, it is vital to not only accurately reconstruct the building’s structure but also to compute thetopology between the objects. More specifically, we target the topology of the reconstructed partial wall geometry as this forms thebasis for other objects.In this work, a novel method is presented to automatically adjust the topology of wall geometry in an as-built BIM. We presenta semi-automated method that procedurally evaluates the configuration of reconstructed objects and adjusts them to create a morefaithful BIM. A wall connection evaluation algorithm is proposed that takes as input the centrelines of partial wall geometry and aset of floor and ceilings mesh segments and outputs the topologically adjusted objects. The method is tested on a variety of scenesand shows promising results to reliably compute the topology of as-built models. The generated geometry is similar to the geometricmodification proposed by expert modelers. A key advantage is that the algorithm operates directly in Revit and Rhino and can beused for new models as well as for updating existing models.

1. INTRODUCTION

The reconstruction of a Building Information Modeling (BIM)of an existing building involves the creation of a set of paramet-ric objects that form a coherent structure according to the builtsituation. First, the structure of the building is created i.e. thewalls and slabs, after which other objects can be added to themodel (Tah et al., 2018). In this work, we focus on the wallssince the slabs can be semi-automatically generated given thewatertight representation of the walls. These objects are createdeither by extruding the 2D plans of the structure or by design-ing it based on point cloud data (Gimenez et al., 2015). Thelatter is preferred as plan information is often inaccurate due toundocumented construction changes. The procedure involvesexperts to manually locate the points that belong to the walls ofthe structure and create a logical BIM from these observations.As this is a tedious and time consuming procedure, the uptakeof automating the process is enormous for the Architectural En-gineering and Construction industry (AEC) (Kavanaugh, 2013,Volk et al., 2014). However, the retrieval of wall geometry frompoint cloud data has several obstacles. First of all, only a portionof the wall faces of each wall is observed. There are occlusionsfrom furniture and false ceilings but also the connections of thewalls cannot be captured with remote sensing. While the geo-metry of each wall can be reasonably accurately reconstructedfrom the observations, the topology and the connections of thewalls can only be derived from reasoning about the building lo-gic of built structures. It is specifically this topology that is theemphasis of this work. We look to automate the reconstructionof the connections between wall elements given a set of wallobservations.∗Corresponding author

The reconstruction of the wall topology is part of what is re-ferred to as the Scan-to-BIM process. It is considered the finalstep to create a faithful BIM representation of a structure frompoint cloud data in an unsupervised manner. Scan-to-BIM cur-rently is still a process that is part of ongoing research (Pat-raucean et al., 2015). Both bottom-up or top-down proceduresare proposed (Hichri et al., 2013). The former considers a rangeof measurements from the site as input and extracts increasinglyhigher level information from the observations until the inten-ded geometry can be reliably constructed. This is a generalapproach which relies on building logic to interpret the data. Itis often used in projects that do not have access to other reliablesources of information such is the case with insufficiently docu-mented buildings and heritage structures. In contrast, the latterrelies on this prior information for the point cloud interpretationand reconstructs a well defined number of objects in a selectivemanner. This is considered a supervised pattern recognition ap-proach which is commonly used in Scan-vs-BIM (Bosche etal., 2013, Bosche et al., 2014). In this research, we propose abottom-up method since it is generally applicable (Gimenez etal., 2015). Furthermore, we solely rely on the point cloud andbuilding logic for the point cloud interpretation since the incor-poration of e.g. sensor information would make the proceduresensor-dependent.

The topology reconstruction is preceded by a number of stepsin the Scan-to-BIM process including the segmentation, classi-fication, clustering and finally the reconstruction of the objects.This research solely discuses the final step and takes as inputthe partial LOD200 BIM walls from prior work (Bassier et al.,2018b, Bassier, Vergauwen, 2019). The presented method takesthese inputs and iteratively computes the most likely connec-

, M. Vergauwen

Commission V, WG V/7 & Commission IV, WG IV/6

The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLII-5/W3, 2019 Capacity Building and Education Outreach in Advanced Geospatial Technologies and Land Management, 10–11 December 2019, Dhulikhel, Nepal

This contribution has been peer-reviewed. https://doi.org/10.5194/isprs-archives-XLII-5-W3-13-2019 | © Authors 2019. CC BY 4.0 License.

13

Figure 1. Automated wall topology adjustment using Rhinocommon and Rhino.Inside: (1) The observed wall geometry from remotesensing shown in Rhino, (2) The partial wall geometry is shown in Revit (grey), (3) the wall axes are shown as green lines and (4) the

additional wall geometry created to complete the topology is shown in green.

tions. The method differs from existing literature as it dealswith both straight edges as well as arcs and polylines. The goalis to compute a watertight model of multi-storey buildings in anautomated manner. Furthermore, our approach creates both Re-vit and Rhino native geometry which ensures data compatibilitywith a wide range of industry applications.

The remainder of this work is structured as follows. The back-ground and related work is presented in Section 2. In Section 3,the methodology is presented. The test design and experimentalresults are proposed in Section 4. Finally, the conclusions arepresented in Section 5.

2. BACKGROUND & RELATED WORK

The reconstruction of the topology between walls is still thetopic of ongoing research (Nguyen, Le, 2013). Both 2D and3D methods are proposed along with a variety of inputs. Firstof all, there are the wall intersection algorithms that computewhich set of wall intersections is the most suitable to representthe as-built BIM. For instance, Valero et al. solves the intersec-tions of pre-segmented wall lines to create a closed area (Valeroet al., 2012). They retain the first intersection which scoreswell for single rooms with proper observations as does Xionget al. (Xiong et al., 2013).

Probably the most commonly used method to create watertightwall geometry is cell decomposition. This is typically a 2Dmethod that relies on a set of semi-infinite rays to compute theboundaries of a target geometry. First, a set of straight linesis computed based on the observed walls. These can repres-ent the wall faces or the centreline of the wall. Following, thisgeometry is used to fit a set of semi infinite rays. The rays in-tersect within the dimensions of the building, thus creating a2D cell grid of the structure. Following, the cells are mergedtogether based on some seeding criteria such as the presence

of floor or ceiling geometry. The rays at the edge of eachcluster subsequently form the boundary of the object’s geo-metry. For instance, Budroni et al. (Budroni, Boehm, 2010)feed their detected wall face planes to a 2D cell decomposi-tion that retains the set of intersections that encloses their floorgeometry. Their method is capable of finding the proper wallintersections of even complex Manhattan world rooms givenunoccluded observations. Previtali et al. (Previtali et al., 2014)and Murali et al. (Murali et al., 2017) use a similar method fortheir detected wall lines. They also use ad-hoc blueprints ofthe building using the floor and ceiling geometry to better eval-uate the potential intersections. Both methods process singlerooms and target the individual wall faces. In contrast, ourmethod targets entire floors and focuses on volumetric walls.Ambrus et al. (Ambrus et al., 2017) successfully extend celldecomposition to entire floors but still targets individual wallfaces in contrast to our method. Another approach that simul-taneously reconstructs multiple rooms is the method of Turneret al. (Turner, Zakhor, 2014). They generate a 2D Delaunaymesh based on their partial wall geometry and use graph-cutsto assign the triangles to different rooms based on a set of seednodes. They then merge the naked edges of each grown patch,which results in a generated floor plan. While their results arepromising for their application, their approach does not oper-ate on volumetric wall objects and they do not attempt to cre-ate watertight geometry. Cell decomposition is also used in avariety of other applications such as roof reconstruction (Kada,McKinley, 2009), building hull extraction, and wall reconstruc-tion (Budroni, Boehm, 2010, Michailidis, Pajarola, 2016). Aninteresting feature is that the topology and geometry is con-structed simultaneously. However, in this research, we considerboth steps separately to compute the best fit partial wall geo-metry before evaluating the topology.

There are several approaches that closely align with our workwhich are discussed in (Ochmann et al., 2019). In comparison



14

to our approach, Mura et al. (Mura et al., 2014, Mura et al.,2016) use 2D cell decomposition on an entire floor and decidewhich intersections are valid based on the floor and ceiling in-formation. Furthermore, they reconstruct volumetric wall en-tities similar to our BIM wall outputs. We extend their workby considering connections between more complex centrelinesof walls such as arcs and polylines. Next, there is Oesau etal. (Oesau et al., 2014). They perform an indoor scene recon-struction based on a 3D cell decomposition. They specificallytarget complex geometry and succeed in creating a watertightmodel of the faces of the interior of buildings. We also oper-ate in 3D but focus on volumetric parametric entities instead ofBREP geometry. In their recent study, Ochmann et al. (Och-mann et al., 2019) proposed a fully automated volumetric re-construction based on 3D cell decomposition, with very prom-ising results. We propose a similar approach but both allowmore types of connections (such as the shortest euclidean dis-tance) and we use more complex centrelines as input. The goalis to not only reconstruct straight walls with proper observa-tions but also deal with the small percentage of complex wallconfigurations.

3. METHODOLOGY

The proposed wall topology reconstruction operates as follows.We start from the partial walls and propose a set of candidateconnections. Based on minimal adjustments, the most logic setof connections is retained. Given the nodes of each connec-tion, the wall centrelines of the walls are adjusted. The methodoperates on IfcWallStandardCase objects, which is the desireddeliverable of many BIM applications. To this end, it was im-plemented in both Revit and Rhino (Fig. 1). The consecutivesteps are discussed in detail in the following paragraphs.

3.1 Preprocessing

As previously stated, several preparatory steps of the presentedapproach are performed in prior work. First, the unstructuredpoint cloud is represented as a voxel octree after which planarpatches are extracted using an efficient parallel processing re-gion growing (Bassier et al., 2017). To reduce the degree ofoversegmentation, a first associative Conditional Random Fieldis employed to merge similar clusters that got separated due todata tiling. Next, the planar patches are subjected to a reas-oning framework that computes class labels for each patch. Apre-trained Random Forests model is used for the classifica-tion (Bassier et al., 2018b). This classification model is chosenfor its robustness against high variance datasets as is the casewith characteristics of building geometry observations due towall detailing, openings, arched walls and so on. The result isa set of labeled segments that replaces the point cloud repres-entation of the building. Following, the segments are clusteredinto groups that represent the individual walls. A second Con-ditional Random Field exploiting local and contextual informa-tion is employed to compute the most likely assignment of thewall segments (Bassier, Vergauwen, 2019). In contrast to theexisting literature, the wall clustering does not solely rely oncoplanarity or parallelity for its clustering, thus again makingit more robust against the high variance of wall observation’scharacteristics. The result is a set of clustered wall mesh seg-ments that represents all the available observations of each wall.This is a highly reliable and accurate representation of the ob-served structure but it is incomplete due to occlusions and thuscan only serve as a static model. Therefore, in a final step, par-tial LOD200 (BIMForum, 2016) wall geometry is established

based on the clustered segments that represents the basic para-metric geometry of the walls (Bassier et al., 2018a). The 2Dcentrelines of these entities serve as the input for the presentedmethod.

3.2 Topology estimation

The estimation of the topology is performed iteratively and con-sists of three main phases (Fig. 2). First, an eligible seed pi isestablished at the end of a centreline. Next, the potential con-nections C are defined for pi based on a set of neighbors Q.Iteratively, the best fit set of connections is withheld until allpossible connections are resolved. In a final step, the new sec-tions of centrelines are created and joined with the partial wallgeometry to form a more faithful BIM. Each step is discussedin detail below.

Seed and neighbors The seed pi and its neighbors Q arefound as follows. The endpoint of a centreline is considereda seed if it does not intersect with another centreline. For a seedpi of a centreline li, the centrelines N are found that contain apoint qi lying within a user defined threshold distance td of pi(Eq. 1). This distance is set equal to the maximum gap a wallis allowed to span over to make a connection and is applicationdependent.

qj =(pi ∈ li ∧ qj ∈ lj : argminqj (‖pi − qj‖)

)Q =

{qj

∣∣∣∀l ∈ L \ {li} ∧ pi ∈ li ∧ qj ∈ lj : ‖pi − qj‖≤ td}

N ={l ∈ L

∣∣∣∀qj ∈ Q : l ∩ qj}

(1)

where the smallest euclidean distance ‖pi − qj‖ is observedto retain a neighboring centreline. Q is the set of the possibleconnection points of pi and N is the set of curves to which Qbelongs.

Candidate connections Given the N and Q, several connec-tions can be made between pi and every qj ∈ Q. In contrastto the literature, we are able to process both straight edges, arcsand polylines. Additionally, we define multiple types of con-nections (Fig. 2): the intersection between the centrelines liand lj , the intersection between the normals at pi and qj andfinally also the connection with the shortest euclidean distance−−→piqj (Eq. 2).

U ={uij

∣∣∣∀lj ∈ N ∧ pi ∈ li : li ∩ lj}

V ={vij

∣∣∣∀qj ∈ Q ∧ pi ∈ li :−→npi ∩ −→nqj

} (2)

where U and V are the intersection points of respectively thefunctions themselves and their normals at pi and qj . This resultsin a set c ∈ C of three types of potential connections betweenpi and its neighboring centrelines (Eq. 3).

c ∈ C =

Intersection: {−−→piuij ,

−−→qjuij}Orthogonal connection: {−−→pivij ,

−−→qjvij}Shortest connection: −−→piqj

(3)



15

(a) First iteration: {−−−→piuij ,−−−→qjuij} is prioritized over {−−→pivij ,

−−−→qjvij} and −−→piqj .

(b) Second iteration: qi = uij , and thus the orthogal connection {−−−→piuij} is prioritized.

Figure 2. Example wall topology reconstruction of a 3 wall configuration depicting the nearby wall (brown) and floor (purple)segments, the centrelines (red), the potential connections C (green lines), the rejected connections (red dotted lines). The blue lines

are the reconstructed centrelines in subsequent iterations to complete the wall topology.

However, not every possible connection should be considered.It is our assumption that possible connections should not in-tersect with floor or ceiling geometry as this violates buildinglogic. Also, the connection should not intersect with wall seg-ments. Therefore, C is conditioned to not intersect with any

extruded floors, ceilings and walls (Eq. 4).

C′ = C \ {floors, ceilings, walls} (4)

Centreline creation The best suited connection is chosen ina pairwise manner. During each iteration, a maximum of 2 c ∈



16

C is selected to connect two centrelines. The two centrelinesare decided based on the most valid potential connections inbetween them. Subsequently, a hierarchical method is employedprioritizing valid intersecting connections {−−→piuij ,

−−→qjuij} overorthogonal connections {−−→pivij ,

−−→qjvij} and finally over the shortesteuclidean distance −−→piqj . When a connection is established, liand lj are extended with the chosen segments. Once the seg-ments are created, the process is repeated for the next pi untilno more eligible seeds can be found. The major advantage ofthe presented method is that it also deals with complex wallintersections with multiple walls and types of centrelines.

3.3 Wall creation

The result of the topology reconstruction is a set of adjustedwall centrelines. Using Rhino.Inside (Robert McNeel & Asso-ciates, 2019), additional native Revit LOD200 walls are con-structed to update the existing model. The result is topologic-ally sound model in frequently used formats that can be used inwide variety of applications.

4. EXPERIMENTS

The proposed algorithm can be successfully used in Revit andRhino using the Rhinocommon API and Rhino.Inside. Fig-ure 1 shows an example of the interface between both soft-ware and the automatically reconstructed topology. Both theestimation of the candidate connections and the topology ad-justment were performed fully automatically. Several examplesare presented to test the performance of the algorithm. The testsinclude 2-wall connections and also more complex 3-wall con-nections for different configuration of lines, arcs and polylines(Table.1). Each case is simulated with centrelines from a num-ber of mesh observations which is the output from our previouswork (Bassier, Vergauwen, 2019). The centrelines representrealistic conditions with near-parallel functions, different con-figurations and different types of functions. The automaticallyreconstructed geometry is compared to the manually designedwalls.

The wall topology reconstruction computed a proper solutionfor most wall scenarios. There are of course the more com-plicated wall configurations that, without the support of addi-tional information, can have multiple outcomes. However, thepresented centrelines closely align with the manually construc-ted topology. Several weaknesses still remain in the method.First of all, without the presence of observed ceiling, floor andother wall geometry, the method’s reliability can be comprom-ised. While prioritizing intersecting connections gives prom-ising results even in occluded scenarios, we cannot know forcertain whether this is in fact the correct topology. A secondaspect is the choice of the seed pi, which currently is basedon the centreline length. This generally leads to a proper wallconfiguration but not necessarily to the proper walltype. Es-pecially in more complex scenarios where a first adjustment isalready computed, there might be confusion about which wall-type should be chosen for the secondary and tertiary wall seg-ments. We state that the model should still be validated by auser to ensure that the proper walltype is chosen for the ad-ditional centrelines. Finally, there is also the assumption offunction continuity for the intersecting curves. As the distancethreshold increases, the extrapolation of functions (especiallyfor arcs and polylines) become increasingly unreliable. Thisproblem is again tied to the occlusions in the model which shouldbe avoided wherever possible.

5. CONCLUSION

This paper presents an semi-automated method to reconstructthe wall topology of Building Information Models from a set ofpartial centrelines that form the basis of parametric wall geo-metry. The method takes as input 2D curves including lines,arcs and polylines originating from previous work and outputsNative Revit and Rhino wall extensions. Given the centrelinesof a set of LOD200 partial walls, the method computes a setof potential connections between the walls including intersec-tions, orthogonal connections and the shortest euclidean dis-tance between the curves. Iteratively, the best fit connection iswithheld by prioritizing intersections over orthogonal and euc-lidean connections. Additionally, the potential connections areconditioned to not intersect with wall, floor and ceiling seg-ments to avoid false positives. The result is a set of wall adjust-ments that form a more faithful as-built BIM representation.

The experiments indicate that the used method is a promisingtopology reconstruction framework. The proposed wall adjust-ments are similar to what human modelers would prefer and themethod deals with complex curves and configurations whichcan significantly speed up the manual modeling process. Addi-tionally, the method can also operate on existing models as longas a set of centrelines is extracted from the existing walls. Themethod currently solely operates on parametric walls based on2D centrelines. In future work, the wall topology will be exten-ded to deal with more complex wall geometry to further expandthe automation of as-built BIM reconstruction.

6. ACKNOWLEDGMENTS

This project has received funding from the European ResearchCouncil (ERC) under the European Union’s Horizon 2020 re-search and innovation programme (grant agreement 779962)and the Geomatics research group of the Department of CivilEngineering, TC Construction at the KU Leuven in Belgium.

REFERENCES

Ambrus, R., Claici, S., Wendt, A., 2017. Automatic RoomSegmentation From Unstructured 3-D Data of Indoor En-vironments. IEEE Robotics and Automation Letters, 2(2),749–756.

Bassier, M., Klein, R., Van Genechten, B., Ver-gauwen, M., 2018a. Ifcwall Reconstruction from Un-structured Point Clouds. ISPRS Annals of Photogram-metry, Remote Sensing and Spatial Information Sciences,IV-2, 33–39. https://www.isprs-ann-photogramm-remote-sens-spatial-inf-sci.net/IV-2/33/2018/.

Bassier, M., M, B., Van Genechten, B., M, V., 2017.Octree-Based Region Growing and Conditional RandomFields. The International Archives of the Photogrammetry,Remote Sensing and Spatial Information Sciences, VolumeXLII-2/W8, 2017 5th International Workshop LowCost3D – Sensors, Algorithms, Applications, XLII(November),28–29.

Bassier, M., Van Genechten, B., Vergauwen, M.,Genechten, B. V., Vergauwen, M., 2018b. Classificationof sensor independent point cloud data of building objectsusing random forests. Journal of Building Engineering, 1–10. https://doi.org/10.1016/j.jobe.2018.04.027.



17

2 Line 3 Line

Line

Arc

Poly

Table 1: Result of the automated topology estimation given the reconstructed centrelines. Each row represents the fitting results for the Line,

Arc and Polyline for a 2 Line and a 3 Line configuration. The brown lines show the wall segments, the surrounding floor and ceiling segments

are shown in purple. The potential connection C are shown as green lines and the final connections are show as blue lines. The green dots are

the collection of intersection points and p and q



18

Bassier, M., Vergauwen, M., 2019. Clustering of WallGeometry from Unstructured Point Clouds Using Condi-tional Random Fields. Remote Sensing, 42(2/W9), 101–108.

BIMForum, 2016. Level of Development Specification.Technical report, BIMFORUM.

Bosche, F., Guillemet, A., Turkan, Y., Haas, C. T., Haas,R., 2014. Tracking the Built Status of MEP Works: Assess-ing the Value of a Scan-vs-BIM System. Journal of Com-puting in Civil Engineering, 28(4), 5014004–5014013.http://ascelibrary.org/doi/abs/10.1061/(ASCE)CP.1943-5487.0000343.

Bosche, F., Turkan, Y., Haas, C., 2013. Tracking the BuiltStatus of MEP Works : Assessing the Value of a Scan-vs .-BIM System. Journal of Computing in Civil Engineering.http://dx.doi.org/10.1016/j.autcon.2014.05.014.

Budroni, A., Boehm, J., 2010. Automated 3D Reconstruc-tion of Interiors from Point Clouds. International Journalof Architectural Computing.

Gimenez, L., Robert, S., Suard, F., Zreik, K., 2015. Auto-matic reconstruction of 3D building models from scanned2D floor plans. Automation in Construction, 63(2016), 48–56. http://dx.doi.org/10.1016/j.autcon.2015.12.008.

Hichri, N., Stefani, C., Luca, L. D., Veron, P., 2013. Re-view of the As-Built Bim Approaches. 3D-ARCH 2013 -3D Virtual Reconstruction and Visualization of ComplexArchitectures, XL-5/W1(February).

Kada, M., McKinley, L., 2009. 3D Building Reconstruc-tion from LIDAR based on a Cell Decomposition Ap-proach. CMRT09: Object Extraction for 3D City Models,Road Databases and Traffic Monitoring - Concepts, Al-gorithms and Evaluation, XXXVIII, 47–52.

Kavanaugh, J. C. K. S. P. E. B., 2013. Client-Guide-to-3D-Scanning-and-Data-Capture. Technical report, Build-ing Information Modelling Task Group.

Michailidis, G. T., Pajarola, R., 2016. Bayesian graph-cutoptimization for wall surfaces reconstruction in indoor en-vironments. Visual Computer, 1–9.

Mura, C., Mattausch, O., Jaspe Villanueva, A., Gobbetti,E., Pajarola, R., 2014. Automatic room detection and re-construction in cluttered indoor environments with com-plex room layouts. Computers & Graphics, 44, 20–32.

Mura, C., Mattausch, O., Pajarola, R., 2016. Piecewise-planar Reconstruction of Multi-room Interiors with Ar-bitrary Wall Arrangements. Computer Graphics Forum,35(7), 179–188.

Murali, S., Speciale, P., Oswald, M. R., Pollefeys, M.,2017. Indoor Scan2BIM: Building information models ofhouse interiors. IEEE International Conference on Intelli-gent Robots and Systems, 2017-Septe.

Nguyen, A., Le, B., 2013. 3D point cloud segmentation :A survey 3D. 2013 6th IEEE Conference, 225–230.

Ochmann, S., Vock, R., Klein, R., 2019. Automatic recon-struction of fully volumetric 3D building models from ori-ented point clouds. ISPRS Journal of Photogrammetry andRemote Sensing.

Oesau, S., Lafarge, F., Alliez, P., 2014. Indoor scene recon-struction using feature sensitive primitive extraction andgraph-cut. ISPRS Journal of Photogrammetry and RemoteSensing, 90, 68–82.

Patraucean, V., Armeni, I., Nahangi, M., Yeung, J., Bri-lakis, I., Haas, C., 2015. State of research in automaticas-built modelling. Advanced Engineering Informatics, 29,162–171.

Previtali, M., Barazzetti, L., Brumana, R., Scaioni,M., 2014. Towards automatic indoor reconstructionof cluttered building rooms from point clouds. IS-PRS Annals of Photogrammetry, Remote Sensing andSpatial Information Sciences, II-5(June), 281–288.http://www.isprs-ann-photogramm-remote-sens-spatial-inf-sci.net/II-5/281/2014/.

Robert McNeel & Associates, 2019. Rhino.Inside.

Tah, J., Oti, A., Abanda, F., 2018. A state-of-the-art reviewof built environment information modelling (BeIM). Or-ganization, Technology and Management in Construction:an International Journal, 9(1), 1638–1654.

Turner, E., Zakhor, A., 2014. Floor Plan Generation andRoom Labeling of Indoor Environments from Laser RangeData. GRAPP, International Joint Conference on Com-puter Vision, Imaging and Computer Graphics Theory andApplications, 1–12.

Valero, E., Adan, A., Cerrada, C., 2012. Automatic methodfor building indoor boundary models from dense pointclouds collected by laser scanners. Sensors (Switzerland),12(12).

Volk, R., Stengel, J., Schultmann, F., 2014.Building Information Modeling (BIM) for ex-isting buildings - Literature review and futureneeds. Automation in Construction, 38, 109–127.http://dx.doi.org/10.1016/j.autcon.2013.10.023.

Xiong, X., Adan, A., Akinci, B., Huber, D., 2013. Auto-matic creation of semantically rich 3D building modelsfrom laser scanner data. Automation in Construction, 31,325–337.



19

semi-automated topology adjustment of partial …

Documents